Going to focus on contemporary high-throughput assays based on sodium bisulfite treatment of DNA

Microarrays

• Illumina 27k
• Illumina 450k
• Human only (mostly)

Sequencing

• Whole-genome bisulfite-sequencing (WGBS/BS-seq/methylC-seq)
• Targeted methods
  • (Enhanced/Extended) Reduced representation bisulfite-sequencing (eRRBS/RRBS)
  • Capture + bisulfite-sequencing (Roche SeqCap Epi system)
• Any organism (mostly)

I can't tell you everything

• 25 minutes
• I don't know everything!

Will tell you:

• What I find useful as a fairly well-experienced user & developer
• What I am most familiar with
• Where to find out more

Powerhouses

• minfi
• methylumi

Pipelines

• ChAMP
• RnBeads

Standard microarray issues

• Failed probes
• Cross-reactive probes
• Background correction
• Colour (dye) bias adjustment
• Normalisation

450k-specific issues

• Type I and II probes are very different
• CpG-SNPs

• Several pre-processing options available in Bioconductor packages and some comparisons published e.g., wateRmelon

Tools

• minfi
  • detectionP()
  • preprocessRaw(), preprocessIllumina(), preprocessQuantile(), preprocessNoob(), preprocessSWAN(), preprocessFunnorm()
  • dropLociWithSnps()
• RUVm (remove unwanted variation) implemented in missMethyl
• CopyNumber450k

\(\beta\)-values vs. \(\mathcal{M}\)-values

\(\beta = \frac{M}{M + U + 100} \in [0, 1]\)

\(\mathcal{M} = \log(\frac{M + 1}{U + 1}) \in [-\infty, \infty]\)

"We recommend using the \(\mathcal{M}\)-value method for conducting differential methylation analysis and including the \(\beta\)-value statistics when reporting the results to investigators." (Du, P. et al. BMC Bioinformatics 11, 587 (2010))

Differentially methylated probes (DMPs)

• For a given probe, are the group-average level(s) of methylation different?
• limma is the workhorse, e.g., minfi::dmpFinder()
• OPINION: You want a pretty good reason not to use a limma-based approach

Differentially methylated regions (DMRs)

• Identify regions with different group-average level(s) of methylation
• OPINION: DMR finding/testing is somewhat ad hoc, but getting better
• E.g., see minfi::blockFinder() with the bumphunter::bumphunter() backend

Still being figured out …

OPINION

• bsseq for whole-genome bisulfite-sequencing
• BiSeq for reduced representation bisulfite-sequencing
• RnBeads for comprehensive WGBS/RRBS pipeline

There are also non-R/Bioconductor options

Pipeline (including most pre-processing)

• Data QC (e.g., FastQC)
• Read trimming (e.g., Trim Galore!)
• Read alignment (e.g., bismark, bwa-meth)
• Methylation calling (e.g., aligner-supplied software, Bis-SNP, methtuple)

Input to Bioconductor

• Some aligner-specific file format with the following data per-sample:

chr  pos M  U
chr7 666 13 2
chr7 685 12 0

• Most done prior to reading data into R
• More that ought to probably be done

Some issues

• CpGs overlapping genetic variants
• Copy number variation
• Normalisation (nothing much yet available)

Yes, they affect DNA methylation data!

• Introduction to batch effects

Methods better developed for microarray methylation data

• sva: General methods for removing batch effects and other unwanted variation in high-throughput experiments.
• missMethyl: Includes RUVadj() and RUVfit() specific to 450k data.

Cell-type heterogeneity

• Jaffe, A. E. & Irizarry, R. A. Accounting for cellular heterogeneity is critical in epigenome-wide association studies. Genome Biol. 15, R31 (2014).
  • minfi::estimateCellCounts()
  • RnBeads: See vignette for details

• Bioconductor contains many packages for analysing DNA methylation data
  • Explore these using BiocViews

Microarray

• Analysis pipeline is well-established
• OPINION: Start with minfi

Sequencing

• More work done outside of R/Bioconductor
• OPINION: Move to Bioconductor for exploratory data analysis, inference, visualisation, and integration.

Technical biases are rife

• Careful experimental design and a skeptical mind are key